Skip to content

[Kernel] OAITritonExperts MXFP4: include SM 12.x in supported device range #41028

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
tonyliu312 wants to merge 1 commit into
base: main
Choose a base branch
from
+12 −6
Open

[Kernel] OAITritonExperts MXFP4: include SM 12.x in supported device range #41028

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 12 additions & 6 deletions vllm/model_executor/layers/fused_moe/experts/gpt_oss_triton_kernels_moe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -651,11 +651,14 @@ def _supports_current_device() -> bool:
cap = p.get_device_capability()
if cap is None:
return False
# (9,0) <= cap < (11,0) covers CUDA SM90 (Hopper), SM100+ (Blackwell)
# and ROCm gfx942/gfx950 (which map to 9.4/9.5).
# (9,0) <= cap < (13,0) covers CUDA SM90 (Hopper), SM100+ (datacenter
# Blackwell), SM120/SM121 (consumer Blackwell — RTX 50-series, GB10
# /DGX Spark) and ROCm gfx942/gfx950 (which map to 9.4/9.5). The lower
# bound is enforced because the Triton MXFP4 kernels rely on
# SM 9.0+ tensor-core / matmul instructions.
if not has_triton_kernels():
return False
return (9, 0) <= (cap.major, cap.minor) < (11, 0)
return (9, 0) <= (cap.major, cap.minor) < (13, 0)

@staticmethod
def _supports_no_act_and_mul() -> bool:
Expand Down Expand Up @@ -1065,11 +1068,14 @@ def _supports_current_device() -> bool:
cap = p.get_device_capability()
if cap is None:
return False
# (9,0) <= cap < (11,0) covers CUDA SM90 (Hopper), SM100+ (Blackwell)
# and ROCm gfx942/gfx950 (which map to 9.4/9.5).
# (9,0) <= cap < (13,0) covers CUDA SM90 (Hopper), SM100+ (datacenter
# Blackwell), SM120/SM121 (consumer Blackwell — RTX 50-series, GB10
# /DGX Spark) and ROCm gfx942/gfx950 (which map to 9.4/9.5). The lower
# bound is enforced because the Triton MXFP4 kernels rely on
# SM 9.0+ tensor-core / matmul instructions.
if not has_triton_kernels():
return False
return (9, 0) <= (cap.major, cap.minor) < (11, 0)
return (9, 0) <= (cap.major, cap.minor) < (13, 0)

@staticmethod
def _supports_no_act_and_mul() -> bool:
Expand Down
Loading